基于赵炳南流派知识图谱的可解释多跳检索增强问答方法研究OA

Objective This paper takes the knowledge graph of Zhao Bingnan's school of traditional Chinese dermatology as a starting point and proposes a knowledge graph-based interpretable multi-hop retrieval enhancement method(KG-IMHRE)to solve complex question-answering challenges in the field of traditional Chinese medicine.Methods By combining subgraph mining with dynamic retrieval query generation strategies,the method guides large language models to perform multi-hop relational reasoning,enhancing the handling of complex question-answering tasks in traditional Chinese medicine while ensuring the interpretability of the reasoning results.Results Based on the Zhao Bingnan school test benchmark,KG-IMHRE achieved objective evaluation scores of 85.2 for multiple-choice questions,45.78 for question-answering tasks,and an overall score of 73.38;the expert score for question-answering tasks was 4.2.Conclusion Experimental results show significant advantages over various mainstream models and retrieval schemes,significantly improving the accuracy and reasoning ability of the question-answering process.